Fix metrics API for eksctl >= 0.201.0 (#86)

* Check that metrics server exists before installing

* Update version and changelog

CHANGELOG.md

@@ -1,5 +1,8 @@
 # Changelog
 
+## Version 1.4.6 - Bugfix release
+- Fix metrics API on eksctl v0.201.0 and later
+
 ## Version 1.4.5 - Bugfix and improvements release
 - Fix Nvidia driver and ALB controller download issue.
 - Package Nvidia driver as fallback for creating GPU nodepools.

plugin.json

@@ -1,6 +1,6 @@
 {
     "id": "eks-clusters",
-    "version": "1.4.5",
+    "version": "1.4.6",
     "meta": {
         "label": "EKS clusters",
         "description": "Interact with Amazon Elastic Kubernetes Service clusters",

python-clusters/create-eks-cluster/cluster.json

@@ -73,11 +73,12 @@
             "description" : "Leave empty for current default of eksctl"
         },
         {
-             "name": "installMetricsServer",
-             "label": "Install metrics server",
-             "type": "BOOLEAN",
-             "mandatory" : true,
-             "defaultValue" : true
+            "name": "installMetricsServer",
+            "label": "Ensure metrics server installation",
+            "type": "BOOLEAN",
+            "mandatory" : true,
+            "defaultValue" : true,
+            "description": "Some versions of eksctl may install the metrics server with the default add-ons."
         },
         {
             "name": "privateCluster",

python-clusters/create-eks-cluster/cluster.py

@@ -13,7 +13,7 @@
 from dku_kube.kubeconfig import setup_creds_env
 from dku_kube.autoscaler import add_autoscaler_if_needed
 from dku_kube.gpu_driver import add_gpu_driver_if_needed
-from dku_kube.metrics_server import install_metrics_server
+from dku_kube.metrics_server import install_metrics_server_if_needed
 from dku_utils.cluster import make_overrides, get_connection_info
 from dku_utils.access import _is_none_or_blank
 from dku_utils.config_parser import get_region_arg, get_private_ip_from_metadata
@@ -297,7 +297,7 @@ def add_vm_to_sg():
             add_gpu_driver_if_needed(self.cluster_id, kube_config_path, connection_info, gpu_taints)
 
         if self.config.get("installMetricsServer"):
-            install_metrics_server(kube_config_path)
+            install_metrics_server_if_needed(kube_config_path)
 
         c = EksctlCommand(args, connection_info)
         cluster_info = json.loads(c.run_and_get_output())[0]

python-lib/dku_kube/metrics_server.py

@@ -5,7 +5,20 @@
 from .kubectl_command import run_with_timeout, KubeCommandException
 
 
-def install_metrics_server(kube_config_path):
+def has_metrics_server(kube_config_path):
+    env = os.environ.copy()
+    env["KUBECONFIG"] = kube_config_path
+    cmd = ["kubectl", "get", "pods", "-n", "kube-system", "-l", "app.kubernetes.io/name=metrics-server", "--ignore-not-found"]
+    logging.info("Checking metrics server presence with : %s" % json.dumps(cmd))
+    out, err = run_with_timeout(cmd, env=env, timeout=5)
+    return len(out.strip()) > 0
+
+
+def install_metrics_server_if_needed(kube_config_path):
+    if has_metrics_server(kube_config_path):
+        logging.info("Metrics server is already deployed on the cluster. Skipping install.")
+        return
+
     try:
         env = os.environ.copy()
         env["KUBECONFIG"] = kube_config_path